AFF1's promoter window is 92.5% AFF1-AS1's - worse than IRX2's 89.2% - and the project's promoter table has no row for the neighbour at all
also called AFF1-AS1 · AFF1 divergent transcript · AFF1 promoter attribution
The 3' partner of this project's index fusion, AFF1, has another gene starting 151 letters away pointing the other direction. The window used to measure AFF1's on-switch is 92.5% that other gene's window - worse than the IRX2 case this beat was sent to investigate - and the other gene is not in the project's table at all, so nobody could have noticed.
The same thing, for a scientist
`AFF1-AS1` (NCBI GeneID **100506746**, `AFF1 antisense RNA 1`, 4q21.3) sits at `chr4:86,924,893-86,934,850` on the **minus** strand; `AFF1` starts at `chr4:86,935,001` on the **plus** strand [E1][E2]. The TSSs are **151 bp** apart, head to head, so the project's 2 kb window for AFF1 (`chr4:86,934,001-86,936,001`) shares **1,849 of 2,000 bp = 92.45%** of its sequence with an equivalent window at AFF1-AS1 [E2]. **That is worse than IRX2/IRX2-DT's 89.2%**, on the gene that gives this project's index fusion its C-terminal half. Two further facts make this the sharper case of the two. First, **`AFF1-AS1` is absent from the project's 19,404-row promoter table** — the table appears to be built from `NM_` transcripts and AFF1-AS1 has only `NR_038841` — so the contamination is invisible to every within-table check, and the project's own census of overlapping rows (14.37%) does not count it [E3]. Second, **`AFF1-AS1` is named nowhere in this graph's 2,857 entries**; the only occurrences anywhere in the repository are an HGNC reference dump and one gene-set log from beat q148 [E3]. As with IRX2, the two gene **bodies are disjoint** (AFF1-AS1 ends at 86,934,850; AFF1 begins at 86,935,001), so the confusion is confined to promoter-scale measurements and a gene-body measurement of AFF1 would be attributable [E2].
The geometry, from two fetched routes
| source | AFF1 | AFF1-AS1 | Δ TSS |
|---|---|---|---|
local hg38 refGene (NM_001166693 / NR_038841) | chr4:86,935,001-87,141,054 + | chr4:86,924,893-86,934,850 − | 151 bp |
NCBI esummary GeneID 4299 / 100506746 | chrstart 86935010 < chrstop 87141038 ⇒ + | chrstart 86934849 > chrstop 86924893 ⇒ − | 161 bp |
The 10 bp difference is an isoform-start difference (NCBI's gene-level span begins at a slightly later transcript), not a coordinate disagreement. Under either, the shared fraction of a 2 kb window is 92.5% or 92.0%.
The window the project actually scores is chr4:86,934,001-86,936,001. Of that, 151 bp at the low end (86,934,001-86,934,152) is inside AFF1-AS1's gene body and is the only sequence an AFF1-AS1 window would not share, and symmetrically 151 bp at the high end is the only sequence unique to AFF1's window. Everything in between — 1,849 bp, including both TSSs — belongs to both.
Why this was invisible, which is the transferable part
The project's promoter table is one row per gene symbol and appears to be restricted to NM_ transcripts: AFF1 is there, AFF1-AS1 is not; CDK9 is there, its 116 bp neighbour MIR2861 is not. So a check of the form "does another row of my table fall inside this window?" returns no for AFF1, and returns no for 7.2 percentage points' worth of the genome (the gap between the 14.37% within-table census and the 21.60% full-annotation census).
An instrument that can only see the rows it already has cannot report the contamination that matters, because the contaminating feature is by construction a feature the table excluded. This is the same shape as rule 89: a test that allows only what it can already reach is not a test of the thing it is named after.
The practical consequence: any attribution audit of this project's promoter windows must be run against the full annotation, not against the window table. The full-annotation census is the one in finding-q153-21-6-percent-....
What this does and does not threaten
It does not threaten anything this graph currently asserts, because this graph holds no promoter-scale occupancy claim about AFF1's own promoter. AFF1 is not in q147's ALL_LOCI and the q145/q147/q151 analyses never scored it. The finding is a standing hazard, recorded before it is paid for rather than after.
It does threaten the obvious next experiment. "Is the KMT2A fusion bound at its own partner's promoter?" is a natural question in this project — autoregulation of AFF1 by KMT2A::AFF1 would be a striking result — and the 2 kb promoter readout cannot answer it, because a positive would be 92.5% attributable to an antisense RNA nobody in this field has characterised. Anyone running that experiment needs base-resolution coverage or a gene-body window from the start.
| absence class | measured-not-curated |
|---|---|
| ncbi gene ID aff1 as1 | 100506746 |
| ncbi gene ID aff1 | 4299 |
| refgene aff1 | chr4:86,935,001-87,141,054 plus strand, NM_001166693 |
| refgene aff1 as1 | chr4:86,924,893-86,934,850 minus strand, NR_038841 |
| gene bodies disjoint | true |
| why absent from table | the table carries one NM_-derived row per gene symbol; AFF1-AS1 has only an NR_ transcript and MIR2861 is not a refGene NM_ gene, so neither can appear |
in project promoter table
{
"AFF1": true,
"AFF1-AS1": false,
"CDK9": true,
"MIR2861": false,
"PAF1": true,
"MED29": true
}measurements — the numbers behind the chart
[
{
"quantity": "promoter_window_overlap_pct",
"subject": "AFF1's 2 kb promoter window against an equivalent AFF1-AS1 window",
"value": 92.45,
"units": "percent of a 2,000 bp window",
"n": 2,
"universe": "hg38 refGene: AFF1 TSS chr4:86,935,001 (+), AFF1-AS1 TSS chr4:86,934,850 (-); project window chr4:86,934,001-86,936,001",
"null": 21.6,
"note": "null is the fraction of all 19,404 project windows that share >=50% with some other gene's window; AFF1 is far into the tail. 92.0% under NCBI's gene-level start (161 bp)"
},
{
"quantity": "tss_separation",
"subject": "AFF1 TSS to AFF1-AS1 TSS, head to head",
"value": 151,
"units": "bp",
"n": 2,
"universe": "hg38 refGene NM_001166693 txStart against NR_038841 txEnd",
"null": null,
"note": "161 bp from NCBI esummary gene-level spans; the difference is an isoform-start difference"
},
{
"quantity": "n_corpus_entries_naming_aff1_as1",
"subject": "occurrences of the string AFF1-AS1 in this graph's node files",
"value": 0,
"units": "entries",
"n": 2857,
"universe": "all node JSON files under data/graph/nodes on 2026-09-24, exact-string grep",
"null": null,
"note": "two occurrences exist elsewhere in the repository - an HGNC dump and a q148 gene-set log - and neither is an entry"
}
]- If anyone proposes targeting AFF1 transcription - a CRISPRi tile, a promoter-directed oligonucleotide, an enhancer deletion - at the promoter, the intervention necessarily hits AFF1-AS1 as well, and AFF1-AS1 is uncharacterised, so the off-target arm of that experiment has no literature to predict it from.
- Conversely, an antisense RNA at a fusion partner's promoter is a candidate mechanism in its own right: if KMT2A::AFF1 drives AFF1-AS1 rather than AFF1, a programme built on 'the fusion autoregulates AFF1' would be chasing the wrong transcript, and the 2 kb readout would look identical either way.
- Is AFF1-AS1 expressed in KMT2A::AFF1 infant ALL, and does it move with AFF1? Nothing in this graph or in the deposits it uses has been queried for it, and a divergent lncRNA at the index fusion's own partner is an obvious thing to have looked at.
- Does the KMT2A fusion occupy its own 3' partner's promoter? The question has never been scored in this project; if it is scored, it must be at base resolution or at gene-body scale, because the 2 kb window cannot answer it.
- How many other rows of the 19,404-row table have an NR_-only neighbour the table structurally cannot see? The 7.2-point gap between the within-table and full-annotation censuses bounds it, but the per-row list of NR_-only contaminants has not been extracted.
- E1established confidence 5 of 5: Multiple independent lines of evidence; treated as settled in the field.database
AFF1-AS1 is NCBI GeneID 100506746, 'AFF1 antisense RNA 1', cytoband 4q21.3, and NCBI esummary returns chrstart 86934849 greater than chrstop 86924893 on NC_000004.12, i.e. the minus strand with its TSS at chr4:86,934,850, while AFF1 (GeneID 4299) returns chrstart 86935010 less than chrstop 87141038, the plus strand.
NCBI Gene esummary for GeneID 100506746 (AFF1-AS1) and 4299 (AFF1), assembly NC_000004.12 · NCBI Gene / E-utilities · NCBI Gene 2026What this evidence cannot show
The gene id was resolved by esearch on AFF1-AS1[sym] AND human[orgn], whose querytranslation read back as 'AFF1-AS1[sym] AND "Homo sapiens"[Organism]' and returned exactly one id. A first attempt used GeneID 100874058 from a guess and that id is COX10-DT, an unrelated gene - recorded here because the wrong id returned a well-formed record that would have been easy to write down. esummary chrstart/chrstop are 0-based and the strand is inferred from their order.
Evidence that argues against this (1)
- AFF1-AS1 is an lncRNA with, as far as this beat searched, no functional characterisation. A sceptic can argue that an uncharacterised antisense annotation is a bookkeeping entity rather than a gene, and that a KMT2A-fusion peak in this window is 'obviously' AFF1's. That argument is not available from the data: the measurement cannot distinguish them, and the assumption about which gene owns a divergent promoter is exactly the assumption this beat exists to question. Trinklein ND et al. 2004
- E2established confidence 5 of 5: Multiple independent lines of evidence; treated as settled in the field.dataset
In hg38 refGene, AFF1 (NM_001166693, plus strand) has its TSS at chr4:86,935,001 and AFF1-AS1 (NR_038841, minus strand) has its TSS at chr4:86,934,850, 151 bp apart head to head, so the project's 2 kb AFF1 window chr4:86,934,001-86,936,001 shares 1,849 bp (92.45%) with an equivalent AFF1-AS1 window; the two gene bodies do not overlap.
q153 per-entry promoter-window sharing census over this graph's gene and protein entries · Iqbal Lab oncofusion knowledge graph, beat Q153 · this project 2026What this evidence cannot show
Computed against the UCSC hg38 refGene snapshot with dataTime 2020-08-17 that the project downloaded in an earlier cycle, cross-checked against NCBI esummary fetched 2026-09-24 (which gives 161 bp and 92.0% from a slightly later gene-level start). The 2 kb window convention is this project's own; the percentage is a property of that convention as well as of the genome.
Evidence that argues against this (1)
- The two gene bodies are disjoint, so this is a promoter-scale problem only. A gene-body measurement of AFF1 - which spans 206 kb - is completely unaffected and would be fully attributable. Reporting the 92.5% without that qualification would overstate the damage. UCSC Genome Browser et al. 2026
- E3established confidence 5 of 5: Multiple independent lines of evidence; treated as settled in the field.dataset
AFF1-AS1 has no row in this project's 19,404-symbol promoter table and no row in the 133-library promoter matrix, and the string AFF1-AS1 appears nowhere in this graph's 2,857 node files - only in an HGNC reference dump and one q148 gene-set log.
q153 audit of the 19,404-row promoter table and a repository-wide grep for AFF1-AS1 · Iqbal Lab oncofusion knowledge graph, beat Q153 · this project 2026What this evidence cannot show
This is an absence claim about this project's own files, not about the literature. It is bounded by an exact-string grep for 'AFF1-AS1' over data/, docs/ and analyses/ on 2026-09-24; a synonym or an older symbol for the same lncRNA would not be found by it, and none was searched for. The promoter-table absence was confirmed by a direct membership test against the loaded matrix, not by grep. absence_class is measured-not-curated: the gene exists in refGene and in NCBI Gene, and this project's pipeline excluded it by transcript-prefix filtering.
Evidence that argues against this (1)
- The expected_absent_terms field here names the very string this entry is about, which means an automated claim-support auditor searching THIS entry would find it. The absence claim is about the rest of the corpus, not about this file, and that distinction has to be made by a reader rather than by a string search. NCBI Gene / E-utilities 2026
AFF1's 2 kb promoter window shares 1,849 of 2,000 bp with AFF1-AS1's; a promoter-scale occupancy claim about AFF1 would be 92.5% unattributable
- related to →KMT2A::AFF1 (MLL-AF4), t(4;11)(q21;q23)
the index fusion's 3' partner sits at a divergent promoter, so 'does the fusion occupy its own partner's promoter' cannot be answered at 2 kb scale
the worked case for why the within-table census (14.37%) understates the full-annotation census (21.60%): the contaminating neighbour has no row
Q153's geometry and GeneID re-derived from independently fetched NCBI, HGNC, UCSC and Ensembl records and confirmed
Q153's GeneID and 92.45% re-derived and confirmed; its two-gene framing extended to three
refutes one clause only - 'the q145/q147/q151 analyses never scored it'. AFF1 is a row of the 19,404-symbol matrix and the score is 20/27; the rest of that entry stands